Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

QAGait: Revisit Gait Recognition from a Quality Perspective

Zengbin Wang, Saihui Hou, Man Zhang, Xu Liu, Chunshui Cao, Yongzhen Huang, Peipei Li, Shibiao Xu

TL;DR

QAGait tackles the challenge of gait recognition in unconstrained environments by introducing a quality-centric pipeline that pre-filters silhouettes with Maximal Connect Area and Template Match, handles nonstandard postures with lean-aware Alignment, and optimizes features using quality-aware losses QACE and QATriplet. The approach yields substantial improvements on outdoor datasets (notably up to 7.3% Rank-1 gains on Gait3D) while remaining compatible with existing backbones and datasets through cost-effective preprocessing. Key contributions include a unified silhouette quality assessment, lean-aware augmentation/alignment, and adaptive margins driven by a Partial Feature Norm quality indicator. The practical impact lies in more robust gait recognition for real-world surveillance and identification tasks, with minimal computational overhead and easy integration into current pipelines.

Abstract

Gait recognition is a promising biometric method that aims to identify pedestrians from their unique walking patterns. Silhouette modality, renowned for its easy acquisition, simple structure, sparse representation, and convenient modeling, has been widely employed in controlled in-the-lab research. However, as gait recognition rapidly advances from in-the-lab to in-the-wild scenarios, various conditions raise significant challenges for silhouette modality, including 1) unidentifiable low-quality silhouettes (abnormal segmentation, severe occlusion, or even non-human shape), and 2) identifiable but challenging silhouettes (background noise, non-standard posture, slight occlusion). To address these challenges, we revisit gait recognition pipeline and approach gait recognition from a quality perspective, namely QAGait. Specifically, we propose a series of cost-effective quality assessment strategies, including Maxmial Connect Area and Template Match to eliminate background noises and unidentifiable silhouettes, Alignment strategy to handle non-standard postures. We also propose two quality-aware loss functions to integrate silhouette quality into optimization within the embedding space. Extensive experiments demonstrate our QAGait can guarantee both gait reliability and performance enhancement. Furthermore, our quality assessment strategies can seamlessly integrate with existing gait datasets, showcasing our superiority. Code is available at https://github.com/wzb-bupt/QAGait.

QAGait: Revisit Gait Recognition from a Quality Perspective

TL;DR

QAGait tackles the challenge of gait recognition in unconstrained environments by introducing a quality-centric pipeline that pre-filters silhouettes with Maximal Connect Area and Template Match, handles nonstandard postures with lean-aware Alignment, and optimizes features using quality-aware losses QACE and QATriplet. The approach yields substantial improvements on outdoor datasets (notably up to 7.3% Rank-1 gains on Gait3D) while remaining compatible with existing backbones and datasets through cost-effective preprocessing. Key contributions include a unified silhouette quality assessment, lean-aware augmentation/alignment, and adaptive margins driven by a Partial Feature Norm quality indicator. The practical impact lies in more robust gait recognition for real-world surveillance and identification tasks, with minimal computational overhead and easy integration into current pipelines.

Abstract

Gait recognition is a promising biometric method that aims to identify pedestrians from their unique walking patterns. Silhouette modality, renowned for its easy acquisition, simple structure, sparse representation, and convenient modeling, has been widely employed in controlled in-the-lab research. However, as gait recognition rapidly advances from in-the-lab to in-the-wild scenarios, various conditions raise significant challenges for silhouette modality, including 1) unidentifiable low-quality silhouettes (abnormal segmentation, severe occlusion, or even non-human shape), and 2) identifiable but challenging silhouettes (background noise, non-standard posture, slight occlusion). To address these challenges, we revisit gait recognition pipeline and approach gait recognition from a quality perspective, namely QAGait. Specifically, we propose a series of cost-effective quality assessment strategies, including Maxmial Connect Area and Template Match to eliminate background noises and unidentifiable silhouettes, Alignment strategy to handle non-standard postures. We also propose two quality-aware loss functions to integrate silhouette quality into optimization within the embedding space. Extensive experiments demonstrate our QAGait can guarantee both gait reliability and performance enhancement. Furthermore, our quality assessment strategies can seamlessly integrate with existing gait datasets, showcasing our superiority. Code is available at https://github.com/wzb-bupt/QAGait.
Paper Structure (46 sections, 7 equations, 6 figures, 8 tables)

This paper contains 46 sections, 7 equations, 6 figures, 8 tables.

Table of Contents

  1. 1 Introduction
  2. 2 Related Work
  3. 2.1 Gait Recognition
  4. 2.2 Image Quality Assessment
  5. 3 Methodology
  6. 3.1 Revisit Gait Recognition Pipeline
  7. 1) Early Data Processing.
  8. 2) Data Augmentation.
  9. 3) Feature Extraction.
  10. 4) Loss Function.
  11. 3.2 Model Overview
  12. 3.3 Unified Silhouette Quality Assessment
  13. 1) Maximal Connect Area.
  14. 2) Template Match.
  15. 3.4 Lean-aware Data Augmentation and Alignment
  16. ...and 31 more sections

Figures (6)

  • Figure 1: The quality difference when advanced from in-the-lab to in-the-wild gait datasets, including: 1) more extremely low-quality silhouettes that are unidentifiable; and 2) higher quality variances due to various and complex environments.
  • Figure 2: Overview of our QAGait and some examples with quality issues. (a) We design a series of cost-effective quality assessment strategies, alignment, and data augmentation for given gait sequences before feeding them into backbone. During feature learning, we introduce a quality indicator and propose two quality-aware loss functions based on CE loss and Triplet loss (i.e., QACE and QATriplet) to improve model adaptability for various quality variances between gait sequences. (b-d) Some examples with diverse quality issues from in-the-wild gait datasets. HPM: Horizontal Pyramid Mapping chao2019gaitset.
  • Figure 3: Top: Examples of our silhouette quality assessment and alignment. Bottom: Our gait template sequence collection and the statistical results of quality assessment.
  • Figure 4: Our quality-aware loss functions facilitate the optimization of trustworthy gait sequences and mitigate the effect of low-quality gait sequences. Left: QACE utilizes Partial Feature Quality as the quality indicator to generate adaptive margin functions. This is then applied to the logits of the model output. Right: QATriplet emphasizes Pairwise Quality as the quality indicator for adaptively adjusting positive and negative pair margins, respectively. BNNeck luo2019bag.
  • Figure A1: Visualization of our gait template sequences. Each view has 30 frames and here we randomly sample a frame for example. (a) 3D virtual humans, including woman and man. (b) gait templates with different heights and views.
  • ...and 1 more figures